Fast cine MRI of the orbit

Transcription

1 OBJECTIVE QUANTIFICATION OF THE MOTION OF SOFT TISSUES IN THE ORBIT 33 3 Fast cine MRI of the orbit Presented in part at the annual meeting of the Radiological Society of North America, Chicago, Illinois, October INTRODUCTION Reproducibility of results implies precise description of methods. The purpose of this chapter is to help others avoid some of the problems that were encountered in the course of this research while developing orbital cine MRI (Magnetic Resonance Imaging) methods, since journals offer but limited space for method descriptions. The Philips Gyroscan NT 1.5T MR scanner running under software release was used in all studies, but the methods should, generally, carry over to any other 1.5T scanner. The programs used to process MR images are available on the associated website at (see below). 2. PATIENT PREPARATION Careful preparation of the patient prior to the scan is important for successful cine MRI. Cine MRI of the orbit requires active co-operation of the patient during the scan. In addition, the physics of magnetic resonance imaging are difficult to explain in comparison to those of a computed tomography (CT) scan. Therefore, patients are frequently fearful of the procedure without explicitly expressing such. Our approach was as follows. The patient was asked to participate in the study after the nature of the study had been explained. He or she then received a brochure explaining the cine MRI examination and containing the informed consent form. Among other things, the brochure stressed that he or she was expected to lie perfectly still during the scan, while gazing only with the eyes, and not with the head. Recently, a picture of the scanning procedure has also been added (Figure 1).

2 34 OBJECTIVE QUANTIFICATION OF THE MOTION OF SOFT TISSUES IN THE ORBIT Figure 1. NT scanner with patient on dolly outside the scanner bore. The head coil is not in place, the headband has not been fixated and the fixation aid has not yet been fitted. A separate appointment was commonly planned at least one week later. For practical reasons, this appointment was usually combined with a visit to the treating ophthalmologist. At this second appointment, the details of the procedure were again explained, and any questions were answered, as far as possible. Usually, the informed consent was then signed. The visit also allowed any contraindications (presence of any magnetic metal within the skull, of any magnetizable metal anywhere else in the head, of a pacemaker, of claustrophobia or a history of psychosis) to be ruled out once more. 3. MR SCANNING Practical MR scanning is by necessity a compromise between image quality, the time patients can hold fixation, economical use of scanner time slots and the convenience of the scanning protocol to MR operators. In general, longer scanning times per fixation period increase image quality. Longer scanner time slots allow more time for setup of fixation aids, operator intervention and longer cine sequences. It is difficult, however, to achieve these goals. Long scanning times increase motion and blinking artifacts, since it is difficult to gaze at a fixed target without blinking for an extended period. Scanner time slots are limited for

3 OBJECTIVE QUANTIFICATION OF THE MOTION OF SOFT TISSUES IN THE ORBIT 35 Figure 2. Snow White machine fixation aid. economical reasons. MR scans were performed by a rotating team of two MRI operators from a pool of about twelve, any one of whom might be on duty at the moment of orbital cine MR scanning. This arrangement precluded elaborate fixation aids, complicated scanning protocols with many options, and extended operator intervention (for example to reposition patients), even though operators were trained for orbital cine MR scanning and a written protocol was available. We performed many experiments to find an optimal orbital cine MR scanning protocol, given the above constraints. Fixation times over twenty seconds were found to result in motion artifacts, while most subjects and patients could fixate up to fifteen seconds with only small motion artifacts. Consequently, fifteen seconds was estimated to be the upper limit to fixation time. Corneal anesthesia was found not to improve scanning quality if the fixation time was below twenty seconds, and was not used. After it had been shown that objective measurement of motion was possible based on sequences of seven to nine volumes (Chapter 4), the minimal sequence length was set at seven. This allowed us to maximize MR efficiency and minimize patient fatigue, since the patient spent only about 20 minutes in the scanner bore. The time slot was set at 30 minutes.

4 36 OBJECTIVE QUANTIFICATION OF THE MOTION OF SOFT TISSUES IN THE ORBIT Figure 3. Brainwash fixation aid. Top: View of fixation aid with ends of optic fibers (all of them illuminated for clarity). Note the optic fibers exiting from the back of the device. Bottom: view of sequencer box (left) connected to a bank of LED s (center) that fits under the fiber optic bank (right). 4. GAZE SEQUENCING Proper gaze sequencing requires the use of fixation marks [41]. The fixation marks could not be left on the inside of the scanner bore, because such marks confused patients that had to undergo functional MRI experiments that are performed in the same scanner; thus, a fixation aid was needed to maximize patient cooperation and minimize operator time for setup. Two different devices were developed and (partially) manufactured by me to serve as fixation aids. The first device was dubbed the Snow White machine (Figure 2) and consisted of a transparent acrylic half-pipe that fits snugly in the scanner bore. On the inside is a row with nine fixation marks indicated with numbers 1-9. This sequence is horizontal and level with respect to the patient s eyes and face in the scanner. The straight-ahead fixation mark is at number 5. The marks are 8º apart if the rotational centers of the eyes are 200mm from it. This 200mm distance was found to be a practical approximation of the average distance of the eyes to the fixation aid. Though extreme gaze positions are attractive, in practice, many patients were unable to gaze. There are also two rows, vertical with respect to the patient s face, one 20º to the left and one at 20º to the right of the straight-ahead fixation mark. The Snow White machine was left transparent to minimize claustrophobia. The other device was dubbed the Brainwash fixation aid (Figure 3). It was developed because some patients had difficulty understanding the meaning and order of the marks in the Snow White machine. It had generally the same distribution of the fixation marks but small lights were used - instead of numbers. These lights were made to light up one by one in a specific color, synchronized to the cine scanning protocol. Synchronizing was triggered by the sounds of the MR gradient coil, recorded with a microphone, since an electronic triggering output is not available on the NT (for medico-legal reasons). The lights were actually the ends of fiber-optic cables

5 OBJECTIVE QUANTIFICATION OF THE MOTION OF SOFT TISSUES IN THE ORBIT 37 Figure 4. Patient being slid into the scanner, the Snow White machine is in place. running from the scanner control room, where they were lit by a sequencer box that contained three-color Light Emitting Diodes (LED s) for every fiber. Three programs were available on the sequencer, selectable by a switch: a program showing a sequence of horizontal white lights, and two programs showing vertical sequences in red and green respectively. The MR operator could start a sequence by the press of a button, and the sequencer then waited for the sound of the activation of the gradient coil before lighting the next light in the sequence. This way, a patient could be instructed to follow the white, green or red lights respectively. Though initial experiments were favorable, the Brainwash turned out to be too cumbersome to set up. In the later studies, it was no longer used, and only the Snow White machine was used. The eyes were not occluded. A choice had to be made between either a much (1/4) reduced image resolution (due to halving of the acquisition time) with occlusion or no occlusion. Otherwise, occlusion would unacceptably have increased scan time and patient fatigue, because all scans would have had to be performed twice. It was found easy to determine the horizontal and vertical fixation angles from the cine MRI scans, so occlusion was not needed for this. In addition, it was found that most patients are usually able to fixate consciously with their best eye. Thus, there were almost no sudden flips of the fixating eye during a sequence. This is important because such flips would have resulted in one double image and one missing image in an already very brief sequence.

6 38 OBJECTIVE QUANTIFICATION OF THE MOTION OF SOFT TISSUES IN THE ORBIT Figure 5. Survey scans. A. coronal plane; B transversal plane; C sagittal plane. The borders of the volume slab that is typically scanned during cine MR scanning are indicated in yellow. The white lines indicate some of the landmarks used to align the patient s head position with the axes of the scanner. In A, the interhemispheric fissure, in B the line connecting the center of both lenses, in C, the line from the globe-optic nerve center to the optic foramen. 5. MR PROTOCOL A head-coil was used in all studies, since a surface-coil was found not to improve results and increase scanning-time [41]. For the earlier studies (Chapters 4 and 6), two-dimensional transversal and sagittal scans were acquired with an acquisition time of 6 seconds per image [146]. After improvements in the NT software (> release 6.0), a faster protocol was developed. This protocol was used in the later studies (Chapters 5, 7, 8), where three-dimensional coronal volumes were acquired with an acquisition time < 16 seconds pre whole volume. Only this last protocol will be covered in detail in this section. Before actual MR scanning, sagittal, coronal and transversal survey scans were made. See Figure 5. These scans were used to check alignment of the orbital axes and to set the scanned volume. The fast volume protocol had the following settings: Gradient scanning sequence (with Turbo Field Echo), a TE (Echo Time) of ms, a TR (Repetition Time) of 9.36 ms, Flip Angle 20 degrees, matrix 256x256, field of view 60%, slice thickness 2.0mm, slice distance 2.0mm. These settings result in anisotropic volumes of 256x256x20 voxels, with a voxel size of 0.82x0.82x2.0mm3, and an acquisition time of 15.6s. Foldover suppression was not used, since it increased scanning time by 60%, but did not improve image quality, as the foldover was located outside the orbit. Three sequences of volumes were acquired using the above settings. One horizontal sequence of 9 volumes with the patient gazing at the horizontal fixation marks, and two vertical sequences of 7 volumes each with the patient gazing at one of two vertical rows. During scanning, the patient was asked to fixate the mark next to the number, and the scanning was started. Between sequences, a brief rest is given to allow the patient to relax. The volume sequences were then saved on Optical Disks (OD). The MR data for each patient were about Megabytes (MB). See Figure 6.

7 OBJECTIVE QUANTIFICATION OF THE MOTION OF SOFT TISSUES IN THE ORBIT DIGITAL DATA COLLECTION The Gyroscan scanning software stores the scanning data in an image format proprietary to Philips (for medico-legal and commercial reasons). The Gyroscan at the UMC Utrecht has no facilities for data transfer or conversion to publicly available imaging formats, such as the Digital Imaging and Communications in Medicine (DICOM) format. (Such an option is marketed by Philips, however). Thus, the imaging data are only available on the Gyroscan terminals and on Easyvision workstations also supplied by Philips. The MR data can be converted to a normal format by exporting the data from the Gyroscan to Compact Disk-Read Only Memories (CD-ROMs) via an Easyvision workstation. The CD-ROM will then contain the volume sequences in the form of separate DICOM 3.0 image files. This method has several drawbacks: each CD-ROM can contain the data of only about 4-5 patients and it takes about three to four hours to write such a CD-ROM. More important, all volume and sequence information is subsequently lost, and the volume sequences are stored flat as separate DICOM image files with non-mnemonic names of the form IM_0699. About one in two CD-ROMs thus obtained were unreadable, and the whole procedure then had to be repeated. The reason was that the CD-ROM writing capability of the Easyvision workstation is not very stable under most software releases if any other processing is done anywhere else on the Easyvision network. In practice, this meant that the CD-ROMs had to be created at night or in the weekends, when the network was relatively idle. For this thesis, about 15 Gigabytes (GB) of MR data were collected using the above methods. 7. IMAGE PROCESSING TOOLS For the image processing programs the reader is referred to the website associated with this thesis at The software (including sources) and documentation including manuals that I have written and that are related to this thesis are available there. Before using them, please observe the copyright and disclaimer notices below or on the website. All programs are in Java and will thus run without modification on any computer system that has browser capabilities. All can be downloaded free of charge, while some of them are available as applets and so do not need to be downloaded. ý FlowJ, a 2-D optical flow estimation and flow visualization package. FlowJ contains the implementations of four optical flow algorithms: Lucas and Kanade, Uras, Fleet and Jepson and Singh. It also implements four different flow field visualization methods: hedgehog, dynamic color mapping (described in this thesis), spotnoise, and dynamic color spotnoise (described in this thesis). The FlowJ interface and implementation was tested in collaboration with more than 500 users.

8 40 OBJECTIVE QUANTIFICATION OF THE MOTION OF SOFT TISSUES IN THE ORBIT ý Flow3J, a package for 3-D optical flow estimation and visualization. Flow3J implements the 3-D optical flow estimation algorithm and the scintillation rendering algorithm developed in Chapter 5. ý VolumeJ, a volume rendering package that I developed in my spare time. Written in Java, it is fully object-oriented and so allows polymorphic rendering of any object that has a classification and an interpolation method: scalar volumes, but also vector volumes. Two examples of polymorphic rendering that are available are the rendering of red, green, blue (RGB) color volumes (as obtained, for example, by confocal color microscopy imaging systems), and also of 3-D flow fields used in Flow3J (see above). It has been interfaced to ImageJ, the Java image processing program written developed by Wayne Rasband at the United States National Institutes of Health and available at and can read, manage and write DICOM 3.0 (also the Philips and Siemens dialects) image formats. The VolumeJ interface and capabilities were developed and tested in collaboration with over 3000 users. Disclaimer The software described above and any extensions or alterations are provided "as is" and without warranty of any kind, express, implied or otherwise, including without limitation, any warranty of merchantability or fitness for a particular purpose. The software is experimental only and has not been designed for, tested or approved for clinical use. Any use of the software is at the user's risk. Any conclusions based on the software or its use are solely the user s. I expressly disclaim any responsibility or liability for any and all adverse medical or legal effects, including personal, bodily, property or business injury, and for damages or loss of any kind whatsoever, resulting directly or indirectly, whether from negligence, intent or otherwise, from any use or disuse of the software, from errors in the software, or from misunderstandings arising from the software itself or its use.

9 OBJECTIVE QUANTIFICATION OF THE MOTION OF SOFT TISSUES IN THE ORBIT 41 Figure 6. Horizontal gaze cine MR volume sequence of 9 volumes. Across: fixation 1-9 (t-axis). Down: volume frames.